CN113194401B - Millimeter wave indoor positioning method and system based on generative countermeasure network - Google Patents

Millimeter wave indoor positioning method and system based on generative countermeasure network Download PDF

Info

Publication number
CN113194401B
CN113194401B CN202110350742.7A CN202110350742A CN113194401B CN 113194401 B CN113194401 B CN 113194401B CN 202110350742 A CN202110350742 A CN 202110350742A CN 113194401 B CN113194401 B CN 113194401B
Authority
CN
China
Prior art keywords
discriminator
generator
generated
sample
real
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202110350742.7A
Other languages
Chinese (zh)
Other versions
CN113194401A (en
Inventor
陈志刚
陈禄
王磊
李文
姬智
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xian Jiaotong University
Original Assignee
Xian Jiaotong University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Xian Jiaotong University filed Critical Xian Jiaotong University
Priority to CN202110350742.7A priority Critical patent/CN113194401B/en
Publication of CN113194401A publication Critical patent/CN113194401A/en
Application granted granted Critical
Publication of CN113194401B publication Critical patent/CN113194401B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
    • H04W4/02—Services making use of location information
    • H04W4/023—Services making use of location information using mutual or relative location information between multiple location based services [LBS] targets or of distance thresholds
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
    • H04W4/30—Services specially adapted for particular environments, situations or purposes
    • H04W4/33—Services specially adapted for particular environments, situations or purposes for indoor environments, e.g. buildings
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W64/00—Locating users or terminals or network equipment for network management purposes, e.g. mobility management
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D30/00—Reducing energy consumption in communication networks
    • Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Signal Processing (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Position Fixing By Use Of Radio Waves (AREA)

Abstract

The invention discloses a millimeter wave indoor positioning method and system based on a generative confrontation network, which comprises the steps of obtaining angle observation data of random positions of a plurality of terminals to generate real samples; inputting the real sample and the generated sample to a discriminator at the same time, and training a neural network of the discriminator; calculating the gradient of the objective function relative to each layer of network parameters by adopting an error BP algorithm and optimizing the gradient; inputting the generated sample into an optimized discriminator, and enabling the output of the generated sample passing through the discriminator to be 1 through an optimized generator; repeatedly and iteratively training the discriminator and the generator to achieve the optimal distinguishing capability of the discriminator for obtaining the real data and the generated data, so that the generated data has the same distribution characteristics as the real data; and taking the parameters of the optimal generator as the estimated AP position to complete millimeter wave indoor positioning. The invention realizes the estimation of the AP position and the terminal position by only using a single AP under the condition of unknown indoor environment, and has the advantages of less required information amount and high positioning precision.

Description

Millimeter wave indoor positioning method and system based on generative countermeasure network
Technical Field
The invention belongs to the technical field of indoor positioning, and particularly relates to a millimeter wave indoor positioning method and system based on a generative countermeasure network.
Background
Indoor positioning methods are roughly classified into two types: a positioning method based on a distance model and based on fingerprint information.
(1) Positioning method based on distance model
The positioning method based on the distance model is to calculate the position of a target point by utilizing the geometric properties of a triangle, and comprises a trilateration positioning method, a triangulation positioning method and the like. The trilateration positioning method is to estimate the position of a terminal according to the distance from the terminal to be measured to a known wireless Access Point (AP). The triangulation positioning method is to estimate the position of the terminal according to the angle relationship between the terminal to be measured and the AP. When at least three APs exist around the terminal to be detected, the position of the terminal can be calculated according to a trilateral or triangular positioning formula.
Common positioning algorithms are: time Of Arrival (TOA), Time Difference Of Arrival (TDOA), Angle Of Arrival (AOA), and Received Signal Strength (RSSI).
Although the positioning method based on the distance model can realize better positioning accuracy, the disadvantages are that:
1. the location of the AP needs to be known. 2. 3 and more APs need to be deployed. 3. The indoor environment is complex, multipath effect can occur in the signal propagation process, the algorithm based on the distance model is easily influenced by multipath, and the robustness is poor.
(2) Positioning method based on fingerprint information
The positioning method based on the fingerprint information mainly comprises the steps of collecting fingerprint data, constructing a corresponding fingerprint database, and finally applying a related algorithm to carry out fingerprint matching to finally obtain the position estimation of the terminal to be detected.
The scheme based on fingerprint information reduces the requirement on physical measurement and improves repeatability, but has the following disadvantages:
1. the functional relationship between the fingerprint and the position is not clear, and the robustness of the fingerprint feature is poor. 2. The positioning accuracy is limited by the density of the fingerprint sampling points. 3. In the fingerprint matching stage, the currently acquired data needs to be compared with the data in the database to find the fingerprint with the highest matching degree, and the calculation cost is high.
The existing indoor positioning methods are classified into various methods, such as Infrared Sensor (Infrared Sensor), Ultrasonic Wave (Ultrasonic Wave), Ultra Wide band (Ultra Wide Bandwidth), and radio Frequency identification (rfid). Although the methods have higher positioning accuracy, special hardware equipment needs to be installed, and some hardware is expensive and not suitable for mass deployment, which limits the popularization of the methods.
The main problem of the positioning method based on the distance model is the problem of positioning accuracy. Under the traditional 2.4GHz and 5GHz frequency bands, the signal bandwidth is only 20MHz, and the accuracy of the distance estimation value is low due to small bandwidth and low time resolution. In addition, positioning methods based on channel information of a physical layer, such as csi (channel state information), cfr (channel frequency response), and the like, are affected by multipath effects, and the obtained information is multipath superposition, which brings certain challenges to extraction of features such as DOA, TDOA, AOA, and the like. In addition, the positioning method based on the distance model needs to deploy 3 or more APs, which is not suitable for most practical indoor environments (such as home environment), and needs to know the specific positions of the APs.
Disclosure of Invention
The technical problem to be solved by the present invention is to provide a millimeter wave indoor positioning method and system based on a generative countermeasure network, aiming at the defects in the prior art, so as to realize high-precision single-AP indoor positioning and indoor map construction under the condition of unknown indoor environment.
The invention adopts the following technical scheme:
a millimeter wave indoor positioning method based on a generative countermeasure network comprises the following steps:
s1, acquiring a plurality of terminal random position angle observation data, and generating corresponding real samples;
s2, inputting the terminal random position vector into a generator, and generating a corresponding generated sample by the generator;
s3, inputting the real sample generated in the step S1 and the generated sample generated in the step S2 into a discriminator at the same time, training a neural network of the discriminator to enable the output of the real sample to be 1 and the output of the generated sample to be 0; calculating the gradient of the objective function relative to each layer of network parameters by adopting an error BP algorithm, finely adjusting global parameters by adopting an iterative gradient rising method, and optimizing a discriminator;
s4, inputting the generated sample into the discriminator optimized in the step S3, and enabling the output of the generated sample passing through the discriminator to be 1 through the generator;
s5, training a discriminator and a generator repeatedly and iteratively, and enabling the discriminator to obtain the optimal distinguishing capability of real data and generated data by optimizing and adjusting discrimination network parameters and generator AP position parameters, wherein the generated data has the same distribution characteristics as the real data; and taking the parameters of the optimal generator as the estimated AP position to complete millimeter wave indoor positioning.
Specifically, in step S1, multipath arrival angle information θ of a plurality of terminal positions is collected in an indoor environment l (p n ) Wherein p is n Representing the random position of the terminal, and l is 1,2 and 3
Figure BDA0003002029440000031
The real sample vector is defined as
Figure BDA0003002029440000032
Specifically, step S2 specifically includes: inputting the random position vector of the terminal into a generator according to the position
Figure BDA0003002029440000033
And is located at a position
Figure BDA0003002029440000034
Is measured at a target terminal of the network
Figure BDA0003002029440000035
And differential departure angle
Figure BDA0003002029440000036
The satisfied relation is generated to correspond to any AP position combination on any plane random position
Figure BDA0003002029440000037
The angle vector data of (a) is as follows:
Figure BDA0003002029440000038
wherein the content of the first and second substances,
Figure BDA0003002029440000041
as the position of the terminal
Figure BDA0003002029440000042
At AP location combination a 1 ,a 2 ,...,a L The next generated sample obtained by the generator.
Further, position
Figure BDA0003002029440000043
And is located at a position
Figure BDA0003002029440000044
Is measured at a target terminal of the network
Figure BDA0003002029440000045
The following relationship is satisfied:
Figure BDA0003002029440000046
。
specifically, in step S3, the iterative gradient ascent method is used to fine-tune the global parameters as follows:
Figure BDA0003002029440000047
Figure BDA0003002029440000048
wherein the content of the first and second substances,
Figure BDA0003002029440000049
connecting the weight of the ith neuron of the m layer and the jth neuron of the m +1 layer by a self-coding network of the discriminator, wherein beta is the learning rate of the gradient ascent algorithm,
Figure BDA00030020294400000410
is the partial derivative, J is the discriminator objective function,
Figure BDA00030020294400000411
is a logistic regression layer weight matrix, omega is an autocorrelation network weight matrix, b is an autocorrelation network bias matrix,
Figure BDA00030020294400000412
is the bias of the jth neuron at the mth layer of the self-coding network.
Further, the objective function of the discriminator is as follows:
Figure BDA00030020294400000413
wherein the content of the first and second substances,
Figure BDA00030020294400000414
representing the output of the discriminator on the true sample,
Figure BDA00030020294400000415
representing the output of the arbiter pair to generate samples.
Specifically, in step S4, the generator parameter a is updated by using the existing iterative gradient descent method 1 ,a 2 ,…,a L The following were used:
Figure BDA00030020294400000416
where L is 1,2, … L, η is the learning rate of the gradient descent algorithm.
Specifically, in step S5, the generator generates angle vector data having the same characteristics as the real data, converges the parameters of the angle vector data to an optimal solution as the real AP position, and estimates the plane position by the minimum cost function search method.
Further, the plane position is estimated by a minimum cost function search method, specifically:
Figure BDA0003002029440000051
wherein the content of the first and second substances,
Figure BDA0003002029440000052
for the terminal position estimate, p n Is the position coordinate of any terminal, and is,
Figure BDA0003002029440000053
is an estimate of the position of the AP,
Figure BDA0003002029440000054
is the terminal position p n About AP location
Figure BDA0003002029440000055
Is detected by the differential angle-of-arrival observations,
Figure BDA0003002029440000056
a set of differential angles of arrival is observed.
Another technical solution of the present invention is a millimeter wave indoor positioning method system based on a generative countermeasure network, comprising:
the real sample module is used for acquiring a plurality of terminal random position angle observation data and generating corresponding real samples;
the generating sample module inputs the terminal random position vector into a generator, and the generator generates a corresponding generating sample;
the discriminator module is used for simultaneously inputting the real sample generated by the real sample module and the generated sample generated by the generated sample module into the discriminator, training the neural network of the discriminator, and enabling the output of the real sample to be 1 and the output of the generated sample to be 0; calculating the gradient of the objective function relative to each layer of network parameters by adopting an error BP algorithm, and finely adjusting global parameters by adopting an iterative gradient rising method to be used as the first optimization of a discriminator;
the optimization module is used for independently inputting the generated samples into the discriminator after the discriminator module is optimized, and the output of the generated samples passing through the discriminator is 1 through the optimization generator;
the positioning module repeatedly and iteratively trains the discriminator and the generator in sequence, and the optimum distinguishing capability of the discriminator for obtaining real data and the generated data is achieved by optimizing and adjusting the discrimination network parameters and the generator AP position parameters, so that the generated data has the same distribution characteristics as the real data; and taking the parameters of the optimal generator as the estimated AP position to complete millimeter wave indoor positioning.
Compared with the prior art, the invention has at least the following beneficial effects:
the invention relates to a millimeter wave indoor positioning method based on a generation type countermeasure network, which is characterized in that the distribution characteristics of a generated sample are close to a real sample through an iterative training discriminator and a generator, so that an AP position estimation value is close to a real position, and AP position estimation is completed; and by a minimum cost function searching method, the terminal position estimation is completed by fully utilizing all AP position estimation values and difference arrival angle observation value information.
Further, the real samples obtained in step S1 include distribution features of angle observations at specific AP positions, so as to train the generative confrontation network.
Further, the generated samples generated by the generator in step S2 are used together with the real samples to train the discriminator.
Further, using the position
Figure BDA0003002029440000061
And is located at a position
Figure BDA0003002029440000062
Is measured at a target terminal of the network
Figure BDA0003002029440000063
The satisfied relationships construct the generator.
Further, the method described in step S3 is used to train the discriminator, so as to gradually improve the discrimination capability of the discriminator for the real sample and the generated sample.
Further, the target function of the discriminator is set to make the output of the real sample gradually approach to 1 in the training process of the discriminator, and the output of the generated sample gradually approaches to 0.
Further, the generator is trained through step S4, so that the distribution characteristics of the generated samples of the generator gradually approach to the real samples, and at the same time, the AP position estimation value also gradually approaches to the real position.
Further, through iterative training of the discriminator and the generator in step S5, the distribution characteristics of the generated samples are substantially the same as those of the real samples, so as to achieve the optimal estimation of the AP position.
Furthermore, the terminal position is estimated by a minimum cost function searching method, and all AP position estimation values and difference arrival angle observation value information are fully utilized, so that the estimation error is smaller.
A millimeter wave indoor positioning method system based on a generative countermeasure network is disclosed, the whole system has two inputs and one output, a real differential arrival angle observed value vector sample is directly input into a discriminator module, a random terminal position is input into a generator to generate a generated sample and then input into a discriminator, and the output is a 01 vector; using position through iterative training of discriminators and generators in positioning modules
Figure BDA0003002029440000071
And is located at a position
Figure BDA0003002029440000072
Of the target terminal of
Figure BDA0003002029440000073
And completing the estimation of the AP position according to the satisfied relation.
In summary, the present invention realizes the estimation of the AP position (including the virtual AP) and the estimation of the terminal position by using only a single AP under the condition that the indoor environment is unknown, and has the advantages of less required information amount and high positioning accuracy.
The technical solution of the present invention is further described in detail by the accompanying drawings and embodiments.
Drawings
FIG. 1 is a millimeter wave indoor positioning model;
FIG. 2 is a block diagram of an AP topology estimation system based on a generative countermeasure network according to the present invention;
FIG. 3 is a block diagram of a generative countermeasure network system;
FIG. 4 is a diagram of a neural network structure of the discriminator in the present invention;
FIG. 5 is a schematic diagram of a simulation environment of the present invention;
FIG. 6 is a diagram illustrating the training result of the neural network according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It will be understood that the terms "comprises" and/or "comprising," when used in this specification and the appended claims, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.
It is also to be understood that the terminology used in the description of the invention herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used in the specification of the present invention and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise.
It should be further understood that the term "and/or" as used in this specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.
Various structural schematics according to the disclosed embodiments of the invention are shown in the drawings. The figures are not drawn to scale, wherein certain details are exaggerated and possibly omitted for clarity of presentation. The shapes of various regions, layers and their relative sizes and positional relationships shown in the drawings are merely exemplary, and deviations may occur in practice due to manufacturing tolerances or technical limitations, and a person skilled in the art may additionally design regions/layers having different shapes, sizes, relative positions, according to actual needs.
Please refer to fig. 1, which shows a basic model of millimeter wave indoor positioning. The method only deploys 1 millimeter wave AP in the room with unknown shape and size, and simulates the real AP by using the mirror image (virtual AP) generated by the real AP and related to the wall, namely, the primary reflection path generated by the real AP is regarded as the direct path generated by the virtual AP. The function completed by the invention can be described as that under the condition that the indoor shape, size and the like are unknown and the AP position is unknown, the multipath arrival angle information acquired by a plurality of terminal random positions is utilized to complete the estimation of the AP topological structure firstly and then complete the estimation of the terminal position.
Please refer to fig. 2, which is a block diagram of an AP topology estimation system based on a generative countermeasure network according to the present invention; fig. 3 is a block diagram of a generative countermeasure network system employed in the present invention.
Based on the fact that high-dimensional multipath angle vector data on a large number of random plane positions have two-dimensional distribution characteristics and the geometric relationship among multipath angle vectors, AP positions and target positions, a Generative adaptive neural Network (GAN) is adopted to learn the two-dimensional distribution characteristics, and therefore estimation of AP geometric topology is achieved:
a typical deep neural network-based discriminator is adopted to learn the characteristics of real multipath angle observation data unsupervised, a multipath angle observation data generator is constructed by utilizing the geometric relation among AP positions, indoor random positions and multipath angle vectors, network parameters and generator AP position parameters are identified through optimization adjustment, and meanwhile, the optimum distinguishing capability of the discriminator for obtaining the real data and the generated data is achieved, and the generated data has distribution characteristics almost the same as the real data. Finally, the optimal generator parameter is the AP position we want to estimate.
The invention relates to a millimeter wave indoor positioning method based on a generative countermeasure network, which comprises the following steps:
s1, collecting multipath arrival angle information theta of a plurality of terminal positions in indoor environment l (p n ) Wherein p is n Representing the random position of the terminal, and l is 1,2 and 3
Figure BDA0003002029440000091
And constructs a true sample vector by definition
Figure BDA0003002029440000092
S2, constructing a generator;
by using analytic geometric properties, the positions of the two positions are obtained
Figure BDA0003002029440000093
And is located at a position
Figure BDA0003002029440000094
Is measured at a target terminal of the network
Figure BDA0003002029440000095
Satisfy the requirement of
Figure BDA0003002029440000096
Therefore, any AP position combination corresponding to any plane random position is generated
Figure BDA0003002029440000097
The angle vector data of (a) is as follows:
Figure BDA0003002029440000098
s3, training a discriminator;
to maintain the consistency of the arbiter input with the generator output data format, the arbiter neural network input is defined as the following 2L-2 dimensional vector based on the angle vector information
Figure BDA0003002029440000101
The AP locations are randomly initialized and a generator is used to generate a number of generated samples of the random locations. The generated sample and the real sample obtained in step S1 are simultaneously input to the discriminator, and the discriminator is trained using a back propagation algorithm.
A typical discriminator objective function in a generative confrontation network is constructed as follows:
Figure BDA0003002029440000102
the output of the true sample is made close to 1 and the output of the generated sample is made close to 0 in order to enable the discriminator neural network to distinguish between the current true sample and the generated sample.
Wherein the content of the first and second substances,
Figure BDA0003002029440000103
representing the output of the discriminator on the true sample,
Figure BDA0003002029440000104
representing the output of the arbiter pair to generate samples.
For the target function, calculating the gradient of the target function relative to each layer of network parameters by adopting an error BP (back propagation) algorithm, and similarly fine-tuning global parameters by adopting an iterative gradient ascending method:
Figure BDA0003002029440000105
this is used as the first optimization training of the arbiter.
S4 training generator
The generated data is input to the discrimination network and the output is expressed as:
Figure BDA0003002029440000106
a typical generator objective function in constructing a generative countermeasure network is as follows:
Figure BDA0003002029440000107
make the corresponding discrimination output value
Figure BDA0003002029440000111
The probability close to 1 is maximized, so that the sample distribution characteristics generated by the generator are close to the real samples, and the effect of falseness is achieved, and the optimized AP position estimation can be closer to the real AP position.
Then, the optimization of the generator parameters is realized by minimizing the objective function, and the generator parameters a are updated by adopting the existing iterative gradient descent method 1 ,a 2 ,…,a L (i.e., each AP position) as follows:
Figure BDA0003002029440000112
where L is 1,2, … L, η is the learning rate of the gradient descent algorithm.
S5, alternately and iteratively training and learning the discriminative neural network parameters and the generator parameters through real data and generated data (repeating the step S3 and the step S4), optimally adjusting the discriminative network parameters and the generator AP position parameters, finally enabling the trainer to achieve the optimal discrimination capability of the real data and the generated data, enabling the generator to generate angle vector data with the same characteristics as the real data, and enabling the parameters to converge to an optimal solution: true AP position, thereby enabling AP position estimation.
And S6, estimating the target position.
Assumed to be located at a plane position p n The terminal can measure at least 2 independent differential arrival angles, and the set of observable differential arrival angles is recorded as
Figure BDA0003002029440000113
Derived by using analytic geometry of cosine function
Figure BDA0003002029440000114
Therefore, the plane position is estimated by a minimum cost function search method, specifically:
Figure BDA0003002029440000115
wherein the content of the first and second substances,
Figure BDA0003002029440000116
for the terminal position estimate, p n Is the position coordinate of any terminal, and is,
Figure BDA0003002029440000117
is an estimate of the position of the AP,
Figure BDA0003002029440000121
is the terminal position p n About AP location
Figure BDA0003002029440000122
Is detected by the differential angle-of-arrival observations,
Figure BDA0003002029440000123
a set of differential angles of arrival is observed.
In another embodiment of the present invention, a millimeter wave indoor positioning system based on a generative countermeasure network is provided, which can be used to implement the millimeter wave indoor positioning method based on the generative countermeasure network, and specifically, the millimeter wave indoor positioning system based on the generative countermeasure network includes a real sample module, a generation sample module, a discriminator module, an optimization module, and a positioning module.
The real sample module acquires a plurality of terminal random position angle observation data and generates corresponding real samples;
the generating sample module inputs the terminal random position vector into a generator, and the generator generates a corresponding generating sample;
the discriminator module is used for simultaneously inputting the real sample generated by the real sample module and the generated sample generated by the generated sample module into the discriminator and training the neural network of the discriminator to enable the output of the real sample to be close to 1 and the output of the generated sample to be close to 0; calculating the gradient of the objective function relative to each layer of network parameters by adopting an error BP algorithm, and finely adjusting global parameters by adopting an iterative gradient rising method to be used as the first optimization of a discriminator;
the optimization module is used for independently inputting the generated samples into the discriminator after the discriminator module is optimized, and the output of the generated samples passing through the discriminator is close to 1 through the optimization generator;
the positioning module repeatedly and iteratively trains the discriminator and the generator in sequence, and the optimum distinguishing capability of the discriminator for obtaining real data and the generated data is achieved by optimizing and adjusting the discrimination network parameters and the generator AP position parameters, so that the generated data has the same distribution characteristics as the real data; and taking the parameters of the optimal generator as the estimated AP position to complete millimeter wave indoor positioning.
In yet another embodiment of the present invention, a terminal device is provided that includes a processor and a memory for storing a computer program comprising program instructions, the processor being configured to execute the program instructions stored by the computer storage medium. The Processor may be a Central Processing Unit (CPU), or may be other general purpose Processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), an off-the-shelf Programmable gate array (FPGA) or other Programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, etc., which is a computing core and a control core of the terminal, and is adapted to implement one or more instructions, and is specifically adapted to load and execute one or more instructions to implement a corresponding method flow or a corresponding function; the processor according to the embodiment of the present invention may be used for operating the millimeter wave indoor positioning method based on the generative countermeasure network, and the method includes:
acquiring a plurality of terminal random position angle observation data, and generating corresponding real samples; inputting the terminal random position vector into a generator, and generating a corresponding generated sample by the generator; inputting the real sample and the generated sample into a discriminator at the same time, training a neural network of the discriminator to enable the output of the real sample to be 1 and the output of the generated sample to be 0; calculating the gradient of the objective function relative to each layer of network parameters by adopting an error BP algorithm, and finely adjusting global parameters by adopting an iterative gradient rising method to be used as the first optimization of a discriminator; inputting the generated sample into an optimized discriminator, and enabling the output of the generated sample passing through the discriminator to be 1 through an optimized generator; training the discriminator and the generator repeatedly and iteratively in sequence, and achieving the optimal distinguishing capability of the discriminator for obtaining real data and the generated data through optimizing and adjusting discrimination network parameters and generator AP position parameters, so that the generated data has the same distribution characteristics as the real data; and taking the parameters of the optimal generator as the estimated AP position to complete millimeter wave indoor positioning.
In still another embodiment of the present invention, the present invention further provides a storage medium, specifically a computer-readable storage medium (Memory), which is a Memory device in a terminal device and is used for storing programs and data. It is understood that the computer readable storage medium herein may include a built-in storage medium in the terminal device, and may also include an extended storage medium supported by the terminal device. The computer-readable storage medium provides a storage space storing an operating system of the terminal. Also, one or more instructions, which may be one or more computer programs (including program code), are stored in the memory space and are adapted to be loaded and executed by the processor. It should be noted that the computer-readable storage medium may be a high-speed RAM memory, or may be a non-volatile memory (non-volatile memory), such as at least one disk memory.
One or more instructions stored in the computer-readable storage medium may be loaded and executed by the processor to implement the corresponding steps of the millimeter wave indoor positioning method based on the generative countermeasure network in the above embodiments; one or more instructions in the computer-readable storage medium are loaded by the processor and perform the steps of:
acquiring a plurality of terminal random position angle observation data, and generating corresponding real samples; inputting the terminal random position vector into a generator, and generating a corresponding generated sample by the generator; inputting the real sample and the generated sample into a discriminator at the same time, training a neural network of the discriminator to enable the output of the real sample to be 1 and the output of the generated sample to be 0; calculating the gradient of the objective function relative to each layer of network parameters by adopting an error BP algorithm, and finely adjusting global parameters by adopting an iterative gradient rising method to be used as the first optimization of a discriminator; inputting the generated sample into an optimized discriminator, and enabling the output of the generated sample passing through the discriminator to be 1 through an optimized generator; training the discriminator and the generator repeatedly and iteratively in sequence, and achieving the optimal distinguishing capability of the discriminator for obtaining real data and the generated data through optimizing and adjusting discrimination network parameters and generator AP position parameters, so that the generated data has the same distribution characteristics as the real data; and taking the parameters of the optimal generator as the estimated AP position to complete millimeter wave indoor positioning.
FIG. 4 is a diagram of a neural network of discriminators used in the present invention. In order to learn the two-dimensional distribution characteristics of high-dimensional multipath angle vector data, the discriminator adopts the structures of a multilayer self-coding neural network and a single-layer logistic regression neural network.
1) Self-coding neural network
The self-coding neural network comprises an input layer (layer 1) and an M-1 hidden layer, and the parameter of the mth layer neural network is (omega) (m) ,b (m) ) Wherein, in the step (A),
Figure BDA0003002029440000151
representing the connection weight coefficient between the ith neuron of the mth layer neural network and the jth neuron of the m +1 layer neural network,
Figure BDA0003002029440000152
represents the offset of the jth neuron of the (m + 1) th layer neural network. The output of the previous layer is the data input of the current layer, in particular the input of layer 1 is the input data signal, i.e.
Figure BDA0003002029440000153
The output of the (m + 1) th layer neural network is
Figure BDA0003002029440000154
Wherein the content of the first and second substances,
Figure BDA0003002029440000155
represents the activation function of each neuron:
Figure BDA0003002029440000156
as shown in the figure, according to the principle of self-coding neural network, the output of the (m + 1) th layer neural network is obtained by reverse transmission decoding
Figure BDA0003002029440000157
The network parameters may be adjusted by minimizing the cost function below
Figure BDA0003002029440000158
Wherein, λ and γ respectively represent weight attenuation parameter and sparsity penalty parameter, and the second term and the third term in the above formula are respectively used for preventing neural network overfitting and increasing neuron sparsity constraint. Then, the existing iterative gradient descent method is adoptedTo update the m-th layer neural network parameter to be (omega) (m) ,b (m) ) The following were used:
Figure BDA0003002029440000159
Figure BDA00030020294400001510
and adjusting parameters of the coding neural network layer by adopting the unsupervised training method.
2) Logistic regression network
Output from multilayer self-coding neural network
Figure BDA00030020294400001511
The weighted vector is
Figure BDA00030020294400001512
The neural activation function is a Sigmoid function, and the output can be expressed as
Figure BDA0003002029440000161
Figure BDA0003002029440000162
For ease of illustration, the output of the logistic regression network may also be written as
Figure BDA0003002029440000163
To enable the discriminator to discriminate the generated data as accurately as possible
Figure BDA0003002029440000164
With real data
Figure BDA0003002029440000165
The target function of a typical discriminator in the constructed generative countermeasure network is shown as the formula (3-6).
FIG. 5 is a schematic diagram of a simulation environment according to the present invention; the simulation is performed in a two-dimensional space of 10 × 8m, and average errors of AP position estimates of 9 × 7 different AP position points are counted as shown in the figure. The average error of the AP position estimation is 0.53m, and the average error of the terminal position calculated by the formula (3-9) is 0.32 m.
Referring to fig. 6, the abscissa represents the number of iterative trainings (one training is one discriminant training and one generator training), and the ordinate represents the AP position estimation error. In the initial stage of iterative training, the AP position estimated value is quickly converged to the error within 2m towards the real AP position, then the error convergence speed is slowed down, and after sufficient times of iterative training, the error is converged to a more ideal result.
In summary, the millimeter wave indoor positioning method and system based on the generative countermeasure network of the present invention utilize the known differential arrival angle information, and realize the estimation of the AP position (including the virtual AP) and the estimation of the terminal position only by using a single AP under the condition that the indoor environment is unknown, and have the advantages of less required information amount and high positioning accuracy.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The above-mentioned contents are only for illustrating the technical idea of the present invention, and the protection scope of the present invention is not limited thereby, and any modification made on the basis of the technical idea of the present invention falls within the protection scope of the claims of the present invention.

Claims (6)

1. A millimeter wave indoor positioning method based on a generative countermeasure network is characterized by comprising the following steps:
s1, acquiring a plurality of terminal random position angle observation data, and generating corresponding real samples;
s2, inputting the terminal random position vector into a generator, and generating a corresponding generated sample by the generator;
s3, inputting the real sample generated in the step S1 and the generated sample generated in the step S2 into a discriminator at the same time, training a neural network of the discriminator to enable the output of the real sample to be 1 and the output of the generated sample to be 0; calculating the gradient of the objective function relative to each layer of network parameters by adopting an error BP algorithm, finely adjusting global parameters by adopting an iterative gradient ascent method, and optimizing a discriminator, wherein the finely adjusting global parameters by adopting the iterative gradient ascent method is as follows:
Figure FDA0003675418800000011
Figure FDA0003675418800000012
wherein the content of the first and second substances,
Figure FDA0003675418800000013
connecting the weight of the ith neuron of the m layer and the jth neuron of the m +1 layer by a self-coding network of the discriminator, wherein beta is the learning rate of the gradient ascent algorithm,
Figure FDA0003675418800000014
is the partial derivative, J is the discriminator objective function,
Figure FDA0003675418800000015
is a logistic regression layer weight matrix, omega is an autocorrelation network weight matrix, b is an autocorrelation network bias matrix,
Figure FDA0003675418800000016
a bias for the jth neuron at layer m of the self-coding network; the objective function of the discriminator is as follows:
Figure FDA0003675418800000017
wherein the content of the first and second substances,
Figure FDA0003675418800000018
presentation judgmentThe output of the discriminator on the real sample,
Figure FDA0003675418800000019
representing the output of the discriminator on the generated sample;
s4, inputting the generated sample into the discriminator optimized in the step S3, and enabling the output of the generated sample passing through the discriminator to be 1 through the generator;
s5, training a discriminator and a generator repeatedly and iteratively, and enabling the discriminator to obtain the optimal distinguishing capability of real data and generated data by optimizing and adjusting discrimination network parameters and generator AP position parameters, wherein the generated data has the same distribution characteristics as the real data; taking parameters of an optimal generator as an estimated AP position to complete millimeter wave indoor positioning, generating angle vector data with the same characteristics of real data by the generator, converging the parameters of the angle vector data to an optimal solution as the real AP position, estimating a plane position by a minimum cost function searching method, and estimating the plane position by the minimum cost function searching method, wherein the method specifically comprises the following steps:
Figure FDA0003675418800000021
wherein the content of the first and second substances,
Figure FDA0003675418800000022
for the terminal position estimate, p n Is the position coordinate of any terminal, and is,
Figure FDA0003675418800000023
is an estimate of the position of the AP,
Figure FDA0003675418800000024
is the terminal position p n About AP location
Figure FDA0003675418800000025
Is detected by the differential angle-of-arrival observations,
Figure FDA0003675418800000026
a set of differential angles of arrival is observed.
2. The method of claim 1, wherein in step S1, multipath angle-of-arrival information θ of multiple terminal positions is collected in indoor environment l (p n ) Wherein p is n Representing the random position of the terminal, and l is 1,2 and 3
Figure FDA0003675418800000027
The real sample vector is defined as
Figure FDA0003675418800000028
3. The method according to claim 1, wherein step S2 is specifically: inputting the random position vector of the terminal into a generator according to the position
Figure FDA0003675418800000029
And is located at a position
Figure FDA00036754188000000210
Is measured at a target terminal of the network
Figure FDA00036754188000000211
And differential departure angle
Figure FDA00036754188000000212
The satisfied relation is generated to correspond to any AP position combination on any plane random position
Figure FDA00036754188000000213
The angle vector data of (a) is as follows:
Figure FDA00036754188000000214
wherein the content of the first and second substances,
Figure FDA00036754188000000215
as the position of the terminal
Figure FDA00036754188000000216
At AP location combination a 1 ,a 2 ,...,a L The next generated sample obtained by the generator.
4. A method according to claim 3, characterised by location
Figure FDA00036754188000000217
And is located at a position
Figure FDA00036754188000000218
Is measured at a target terminal of the network
Figure FDA00036754188000000219
The following relationship is satisfied:
Figure FDA00036754188000000220
。
5. the method of claim 1, wherein in step S4, the generator parameter a is updated by using an existing iterative gradient descent method 1 ,a 2 ,…,a L The following were used:
Figure FDA0003675418800000031
where L is 1,2, … L, η is the learning rate of the gradient descent algorithm.
6. A millimeter wave indoor positioning method system based on a generative countermeasure network is characterized by comprising the following steps:
the real sample module is used for acquiring a plurality of terminal random position angle observation data and generating corresponding real samples;
the generating sample module inputs the terminal random position vector into a generator, and the generator generates a corresponding generating sample;
the discriminator module is used for simultaneously inputting the real sample generated by the real sample module and the generated sample generated by the generated sample module into the discriminator, training the neural network of the discriminator, and enabling the output of the real sample to be 1 and the output of the generated sample to be 0; calculating the gradient of the objective function relative to each layer of network parameters by adopting an error BP algorithm, finely adjusting the global parameters by adopting an iterative gradient ascent method, and performing first optimization of the discriminator, wherein the finely adjusting the global parameters by adopting the iterative gradient ascent method is as follows:
Figure FDA0003675418800000032
Figure FDA0003675418800000033
wherein the content of the first and second substances,
Figure FDA0003675418800000034
connecting the weight of the ith neuron of the m layer and the jth neuron of the m +1 layer by a self-coding network of the discriminator, wherein beta is the learning rate of the gradient ascent algorithm,
Figure FDA0003675418800000035
is the partial derivative, J is the discriminator objective function,
Figure FDA0003675418800000036
is a weight matrix of the logistic regression layer, and omega is an autocorrelation network weightA matrix of values, b is a self-encoding network bias matrix,
Figure FDA0003675418800000037
a bias for the jth neuron at layer m of the self-coding network; the objective function of the discriminator is as follows:
Figure FDA0003675418800000038
wherein, the first and the second end of the pipe are connected with each other,
Figure FDA0003675418800000039
representing the output of the discriminator on the true sample,
Figure FDA00036754188000000310
representing the output of the discriminator on the generated sample;
the optimization module is used for independently inputting the generated sample into the discriminator after the discriminator module is optimized, and the output of the generated sample passing through the discriminator is 1 through the optimization generator;
the positioning module repeatedly and iteratively trains the discriminator and the generator in sequence, and the optimum distinguishing capability of the discriminator for obtaining real data and the generated data is achieved by optimizing and adjusting the discrimination network parameters and the generator AP position parameters, so that the generated data has the same distribution characteristics as the real data; taking parameters of an optimal generator as an estimated AP position to complete millimeter wave indoor positioning, generating angle vector data with the same characteristics of real data by the generator, converging the parameters of the angle vector data to an optimal solution as the real AP position, estimating a plane position by a minimum cost function searching method, and estimating the plane position by the minimum cost function searching method, wherein the method specifically comprises the following steps:
Figure FDA0003675418800000041
wherein the content of the first and second substances,
Figure FDA0003675418800000042
for the terminal position estimate, p n Is the position coordinate of any terminal, and is,
Figure FDA0003675418800000043
is an estimate of the position of the AP,
Figure FDA0003675418800000044
is the terminal position p n About AP location
Figure FDA0003675418800000045
Is detected by the differential angle-of-arrival observations,
Figure FDA0003675418800000046
a set of differential angles of arrival is observed.
CN202110350742.7A 2021-03-31 2021-03-31 Millimeter wave indoor positioning method and system based on generative countermeasure network Active CN113194401B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110350742.7A CN113194401B (en) 2021-03-31 2021-03-31 Millimeter wave indoor positioning method and system based on generative countermeasure network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110350742.7A CN113194401B (en) 2021-03-31 2021-03-31 Millimeter wave indoor positioning method and system based on generative countermeasure network

Publications (2)

Publication Number Publication Date
CN113194401A CN113194401A (en) 2021-07-30
CN113194401B true CN113194401B (en) 2022-08-09

Family

ID=76974282

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110350742.7A Active CN113194401B (en) 2021-03-31 2021-03-31 Millimeter wave indoor positioning method and system based on generative countermeasure network

Country Status (1)

Country Link
CN (1) CN113194401B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114826772B (en) * 2022-05-30 2024-03-08 中国联合网络通信集团有限公司 Data integrity verification system

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109239661A (en) * 2018-09-18 2019-01-18 广西大学 A kind of RFID indoor locating system and algorithm based on depth Q network
CN112312541A (en) * 2020-10-09 2021-02-02 清华大学 Wireless positioning method and system
KR20210030133A (en) * 2019-09-09 2021-03-17 홍익대학교 산학협력단 Method for predicting human mobility route based on a generative adversarial network

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107832834B (en) * 2017-11-13 2020-02-14 合肥工业大学 Method for constructing WIFI indoor positioning fingerprint database based on generation countermeasure network
CN109145958B (en) * 2018-07-27 2019-11-08 哈尔滨工业大学 A kind of real scene wisp detection method generating confrontation network based on multitask
US10325201B1 (en) * 2019-01-31 2019-06-18 StradVision, Inc. Method and device for generating deceivable composite image by using GAN including generating neural network and discriminating neural network to allow surveillance system to recognize surroundings and detect rare event more accurately
EP3786864A1 (en) * 2019-08-27 2021-03-03 Siemens Healthcare GmbH Combined indoor and outdoor tracking using machine learning

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109239661A (en) * 2018-09-18 2019-01-18 广西大学 A kind of RFID indoor locating system and algorithm based on depth Q network
KR20210030133A (en) * 2019-09-09 2021-03-17 홍익대학교 산학협력단 Method for predicting human mobility route based on a generative adversarial network
CN112312541A (en) * 2020-10-09 2021-02-02 清华大学 Wireless positioning method and system

Also Published As

Publication number Publication date
CN113194401A (en) 2021-07-30

Similar Documents

Publication Publication Date Title
Jang et al. Indoor localization with WiFi fingerprinting using convolutional neural network
CN107071743B (en) Rapid KNN indoor WiFi positioning method based on random forest
Zhang et al. Deep neural networks for wireless localization in indoor and outdoor environments
Zhang et al. DeepPositioning: Intelligent fusion of pervasive magnetic field and WiFi fingerprinting for smartphone indoor localization via deep learning
Bae et al. Large-scale indoor positioning using geomagnetic field with deep neural networks
CN107727095B (en) 3D indoor positioning method based on spectral clustering and weighted back propagation neural network
CN106851571B (en) Decision tree-based rapid KNN indoor WiFi positioning method
US5537511A (en) Neural network based data fusion system for source localization
CN111901749A (en) High-precision three-dimensional indoor positioning method based on multi-source fusion
BelMannoubi et al. Deep neural networks for indoor localization using WiFi fingerprints
CN111461251A (en) Indoor positioning method of WiFi fingerprint based on random forest and self-encoder
CN113194401B (en) Millimeter wave indoor positioning method and system based on generative countermeasure network
Chen et al. A wifi indoor localization method based on dilated cnn and support vector regression
Yang et al. Multi-floor indoor localization based on RBF network with initialization, calibration, and update
Wei et al. RSSI-based location fingerprint method for RFID indoor positioning: a review
CN111263295B (en) WLAN indoor positioning method and device
Li et al. Sea/land clutter recognition for over-the-horizon radar via deep CNN
Ding et al. Microphone array acoustic source localization system based on deep learning
CN117098067A (en) Multi-mode deep learning indoor positioning method based on gradient fusion
WO2021103027A1 (en) Base station positioning based on convolutional neural networks
Dai et al. Indoor positioning algorithm based on parallel multilayer neural network
Wisanmongkol et al. An ensemble approach to deep‐learning‐based wireless indoor localization
CN114051209B (en) Fingerprint positioning method based on intelligent reflecting surface and scene geometric model
CN115908547A (en) Wireless positioning method based on deep learning
CN113645565B (en) Indoor positioning method based on hexagonal closest packing structure

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant